Oncology trials are consistently among the hardest to enroll. Eligibility criteria tend to be longer and more specific than in other indications. The patient populations are often heavily pre-treated, which creates complex prior therapy exclusions. And the competitive landscape for patients at major cancer centers means that a patient who qualifies for your trial may also qualify for two others.
What is less often discussed is how much performance varies within oncology, across trials that look structurally similar. Industry benchmarking data, where it exists, typically aggregates across therapeutic areas and trial phases in ways that obscure the within-indication spread. From what we observe across our work with sites, the enrollment rate gap between median and top-quartile performers in comparable oncology trials is frequently in the range of three to five times. Understanding what drives that gap matters more than citing a single average figure.
What "Enrollment Rate" Actually Measures
Before diving into what separates high-performing sites, it is worth being precise about what enrollment rate means and where it can mislead. Enrollment rate is typically measured as patients enrolled per site per month. That number conflates several distinct things: the size of the eligible patient population at that site, how aggressively the site is identifying candidates, screen failure rate once formal screening begins, and dropout between consent and randomization.
Two sites can have the same enrollment rate for completely different reasons. One site might be pulling from a genuinely larger eligible population and enrolling at a low yield. Another might be working from a smaller pool but identifying candidates more systematically and losing fewer to screen failure. The same headline number, very different operational realities.
This distinction matters because the interventions that help each of those sites are not the same. If the problem is candidate identification, better pre-screening tooling helps. If the problem is screen failure rate, the issue is more likely protocol criteria design or how eligibility questions are being interpreted at the site level. Benchmarks that aggregate these into a single rate point are useful for trending but not for diagnosis.
Where High-Performing Sites Diverge
From our conversations with site coordinators and from the patterns we see in how protocols get worked up at the sites we support, a few consistent differentiators emerge in oncology.
The first is treatment history documentation. Oncology protocols almost always include prior therapy exclusions: patients who have received agent X, or any agent in class Y, within some defined window are excluded. High-performing sites have systematic ways of capturing that history. This sounds basic, but in practice it requires either a well-maintained internal tumor board registry, a structured EHR field that gets reliably updated, or a coordinator with the time and training to abstract notes from outside records. Sites that lack one of those three typically have spotty prior therapy data, which means more time per candidate and more screen failures when the gap surfaces during formal screening.
The second differentiator is how sites handle patients who are close to the boundary of an exclusion criterion. Almost every complex oncology protocol has at least a handful of criteria that require clinical judgment to apply: prior therapy washout periods that do not align cleanly with treatment dates in the record, lab values from a result that may be outdated, or a diagnosis that appears in one note but not in the problem list. Sites that have run oncology trials before have a developed sense of which criteria need escalation to the PI and which can be evaluated by a sub-investigator. Sites working through that judgment framework for the first time are slower, not because of capability differences, but because each of those decisions is being made fresh.
The third is referral network density. Top-performing oncology sites in our experience are not just pulling from their own patient population. They have relationships with community oncologists in their region who refer patients for trial consideration. Building that network takes time that predates any individual trial. Sites that have invested in it can access a meaningfully larger effective patient pool than their own practice volume would suggest.
The Screen Failure Problem in Oncology
Screen failure rates in oncology are notably higher than in many other indications. Published data across multiple sources consistently suggests oncology screen failure rates in the range of 40 to 60 percent for later-line solid tumor trials, with some precision oncology studies running even higher. That is not a data point we are asserting with a specific citation, but it is consistent with what we observe in the work we do.
The interesting question is where in the screening process those failures are occurring. Early screen failures, at the pre-screening stage before the patient has been formally consented, represent sunk coordinator and PI time but no patient burden. Late screen failures, after consent and during the formal screening period, are more costly on every dimension: wasted patient time, wasted site resources, and timeline impact.
High-performing sites push failures earlier in the funnel. They do more work up front to rule out patients who are unlikely to pass formal screening before engaging them. That requires better upfront information about prior therapy history, comorbidities, and performance status, which brings us back to the pre-screening tooling and workflow questions.
We are not saying that getting screen failures to zero is a realistic goal in oncology. Many screen failures reflect information that genuinely did not exist before the formal screening labs or imaging. But a meaningful fraction of them reflect information that was available in the record and simply was not surfaced during pre-screening. Closing that gap is where the performance difference tends to live.
Protocol Design Choices That Shape Enrollment Speed
There is a category of enrollment benchmark variation that no site-level intervention can address: the variability introduced by protocol design choices themselves.
The number of exclusion criteria is the most studied dimension of this. More criteria means a smaller eligible fraction of the population and more work per candidate to determine eligibility. But it is not the only relevant factor. The operational complexity of individual criteria matters at least as much. A single criterion requiring chart abstraction of prior biologic therapy across a five-year window is more enrollment-limiting than three criteria that can be checked from a structured lab result.
Temporal restriction criteria, where eligibility depends on the timing of prior events relative to enrollment, require precise date data that is often incomplete in EHRs. If a patient received their last chemotherapy cycle at another institution and those records are not in the site's system, verifying washout compliance requires a medical records request, which adds days to the pre-screening timeline per patient.
None of this is a reason to weaken eligibility criteria that exist for legitimate scientific reasons. But the enrollment rate a trial achieves is partly a function of design choices that get made well before the first site is activated. Benchmarks that do not account for protocol complexity as a covariate are limited in what they can tell you about site performance versus trial design performance.
Using Benchmarks Without Over-Interpreting Them
Enrollment benchmarks are most useful as red flags, not targets. If a site is enrolling at less than half the rate of comparable sites on a comparable protocol, that is a signal worth investigating. If a site is hitting the median on a protocol with unusually complex criteria, that might actually represent strong operational performance that the raw number obscures.
The work of understanding oncology enrollment performance requires getting underneath the aggregate rate to the component parts: candidate identification yield, pre-screening time per patient, screen failure rate and timing, and conversion from consent to randomization. Each of those components has different levers. Working from a single enrollment rate metric tends to produce interventions that address the symptom rather than the cause.